Turn domain knowledge into Claude Code skills — NotebookLM research → structured SKILL.md generation → automated validation pipeline.
Manually writing SKILL.md files is slow: research takes hours, hallucinations creep in, and iteration is painful. This skill orchestrates a 4-phase pipeline that combines NotebookLM's source-grounded research with Claude Code's execution:
User Intent → NotebookLM Source Ingestion → SKILL.md Extraction → Validation → Test & Iterate
- No hallucinations — NotebookLM answers only from sources you provide, never invents
- Minutes, not hours — What used to take a full afternoon now takes a few minutes
- Structured output — Extracts production-ready SKILL.md with proper YAML frontmatter
- Built-in iteration — Test failures feed back into NotebookLM for automatic fixes
# One-time setup
notebooklm login # Google OAuth (opens browser once)/create a skill for high-conversion landing page copywriting
The skill will:
- Ask you for source materials (PDFs, URLs, YouTube)
- Create a dedicated NotebookLM notebook and index them
- Extract a complete, validated SKILL.md
- Test it and iterate if needed
clawhub install notebooklm-skill-factorygit clone https://github.com/kidcvs/notebooklm-skill-factory.git
cp -r notebooklm-skill-factory ~/.claude/skills/Or for Hermes:
cp -r notebooklm-skill-factory ~/.hermes/skills/├── SKILL.md # Main skill definition
├── scripts/
│ └── parse-skill-output.py # Parse NotebookLM JSON → SKILL.md
└── references/
└── skill-extraction-prompt.md # Prompt templates for extraction
| Platform | Status | Install |
|---|---|---|
| Claude Code | ✅ | ClawHub / Manual |
| OpenClaw | ✅ | ClawHub / Manual |
| Hermes | ✅ | Manual |
- NotebookLM CLI —
notebooklmcommand - Python 3.10+ (stdlib only, no pip packages)
skill-creatorandskill-vetterskills (for validation phase)
MIT — see LICENSE